Reader error

[wand3r3r]

Title fail. removing the article...http://www.theinquirer.net/inquirer...rst-128gb-supercomputing-server-graphics-card

 

[Hitman928]

Here we are discussing 2 GB vs. 3/4/6 GB. http://www.theinquirer.net/inquirer...rst-128gb-supercomputing-server-graphics-card

it's 12, not 128

another reason theinquirer should never be used as a source.

 

[joshhedge]

Here we are discussing 2 GB vs. 3/4/6 GB. http://www.theinquirer.net/inquirer...rst-128gb-supercomputing-server-graphics-card

If you actually read the article you would have noticed they mentioned 12GB throughout.

Toms article - http://www.tomshardware.com/news/amd-firepro-s10000-12gb,25083.html

 

[f1sherman]

And then you remember that it's a dual GPU...

We started with 128GB and now we're down to 6GB

 

[Ferzerp]

And then you remember that it's a dual GPU...

We started with 128GB and now we're down to 6GB

So basically, it's gone from a title claiming a "first" to being the same memory as another product released a 5 months ago (K6000)?

 

[f1sherman]

So basically, it's gone from a title claiming a "first" to being the same memory as another product released a 5 months ago (K6000)?

Or to be exact - half of usable VRAM compared to K6000.

6 vs 12GB ^_^

 

[Ferzerp]

I'm assuming that for compute, the memory need not be mirrored for all GPUs. For sli/cfx yes, but compute there is no reason to do that. Treat them as discrete gpus with dedicated memory per.

 

[f1sherman]

Aye, you can fit 12GB of data in compute tasks; 6GB per GPU.
Which does not give you exactly the same level of comfort, even in compute tasks, as having 1 GPU with 12GB

As a workstation card, it's good old CF mirroring, hence 6GB.